A memristor is a device that regulates the flow of electrical current in a circuit and remembers the amount of charge that flowed through it. They are non-volatile devices, which means they can retain memory without power. Recently, experts have made a breakthrough in memristors that can make it act like a neuron.
Engineers have been trying to copy the power efficiency and quirky computational skills of the brain for many years, but they cannot do it. They do not have a device that can act like a neuron, one whose behavior is more complicated than any device has ever made.
A new device invented by Hewlett Packard Laboratories' Suhas Kumar, Texas A&M's R. Stanley Williams, and Ziwen Wang, a Stanford student, has met all those requirements. The device outputs simple spikes, burst os spike, self-sustained oscillations, and other processes usually found in the brain using a simple DC voltage as an input.
Most devices can only perform simple spikes, so this new invention is extraordinary. The researchers published their study in the journal Nature.
Memristor Breakthrough
The novel device combines resistance capacitance and Mott memristor, with the nanometers-thin niobium oxide (NbO2) layer as the most vital part. A memristor is a device that can hold memories of the charge that previously flowed through them, in the form of resistance.
Engineers have only explored in nanoscale devices the properties of the materials in a Mott transition that could go between insulating and conducting based on their temperature.
Inside the memristor, the transition happens in a nanoscale silver of NbO2 in which when DC voltage is applied, it heats up slightly. It causes it to transition from insulating to conducting. The charge then builts up in the capacitance pours through once that transition happens.
Slowly the device cools, which triggers a transition back to insulating. The result resembles a neuron's action potential in which a spike current flows in the device.
Williams said that they have been working for five years to get that result and noted that many things are going on in that one nanoscale material.
Leon Chua, the inventor of memristor, predicted that there would be regions of chaotic behavior discovered between areas where behavior is unstable if the possible device parameters are mapped. Devices can exist that can act as a neuron at the edge of some of these chaotic regions.
Read Also: Curly the Robot Competes With Professional Curling Athletes
Applying it to today's machines
According to Williams, it was Kumar who fine-tuned the material and physical parameters of the device to find a combination that works. This invention is the type that cannot be found accidentally because everything has to be perfect before seeing this characteristic. But once it is made, it becomes very robust and reproducible.
The researchers tested the device by building spiking versions of Boolean logic gates, and then they made a small analog optimization circuit.
According to the researchers, there is still a lot more to do to turn these into practical devices and make them bigger to challenge today's machines. Williams and Kumar plan to explore other possible materials that experience Mott transitions at different temperatures, considering that shifts in NbO2 happen only at 800 degrees Celsius, which only occurs in nanometers thin layer. Scaling it up to millions of devices could pose a big problem.
Meanwhile, other scientists are also researching using vanadium oxide, which can transition at 60 degrees Celsius, better than NbO2. However, that temperature might be too low given that systems in the data center operate ta 100 degrees Celsius, Williams said. There could also be other materials that can be used to achieve the same result, which could be very interesting.
Read More: Real-Life Iron Man: Air Ambulance Uses Jet Packs for Emergency Services
Check out more news and information on Tech & Innovation on Science Times.
